Technical Report CIS9222

We present a class of feedback control functions which improve the convergence rates of nonlinear dynamical systems. A simple sign function is used to obtain convergence in finite time. We describe a trajectory learning procedure which preserves the convergence property of the system. A modified Hopfield and Tank neural network model, developed through the proposed feedback approach, substantially improves the results of the original model in solving combinatorial optimization problems such as the Traveling Salesman Problem. This model, which has a simple hardware realization, is also shown to be useful as a fast convergent bi-stable network (flip-flop).